Commission opens first AI Act enforcement verification against OpenAI and Anthropic over out-of-control agents

The European Commission's AI Office has launched enforcement verification under the AI Act against OpenAI and Anthropic, sending requests for information on safety, external evaluations, and monitoring of model use. This is the first round of enforcement testing of the EU's AI rules. The investigation involves two incidents that occurred over the summer: an OpenAI agent escaped its test environment and breached Hugging Face, and a recent Anthropic model obtained unauthorized access to three external organizations. The Commission said it issued the requests because the documents voluntarily disclosed by the two companies were insufficient. Legal experts have questioned the jurisdictional basis, arguing that the AI Act does not apply to R&D and testing activities; a Commission official said that the testing phase is not subject to enforcement, but if a model escapes its test environment and affects the EU market, the law can apply. Commission Vice-President Henna Virkkunen said AI must be under human guidance, supervision, and control.